Artificial intelligence is making the stock markets boom and helping Trump’s popularity. Famous investors are now pointing out a question that should only interest accountants.
12/01/2025, 10:09 p.m12/01/2025, 10:09 p.m
Niklaus Vontobel / ch media
Donald Trump had made big promises for his second term in office: “Incomes are skyrocketing, inflation is disappearing completely, the middle class is prospering like never before.” However, according to consumers, he has not kept these promises, as a survey by the University of Michigan shows. Apart from the Corona period, their mood is worse than it has been in 40 years.
US President Donald Trump: If the AI boom crashes, will it crash too? Image: AI generated/Shutterstock AI
Trump’s popularity ratings are falling; recently the trend was almost reminiscent of a free fall. “No president in recent history has fallen so far so quickly as Trump,” writes The Economist magazine. And Trump would fall even faster if it weren’t for the boom in artificial intelligence (AI). According to one estimate, gigantic investments in data centers and superchips were responsible for 92 percent of economic growth in the first half of the year. Without the AI boom, the economy would have come to a standstill.
However, this boom is shakier than the hype would suggest. It even stands or falls on a puzzle that should actually only interest accountants: How quickly do the data centers and superchips have to be written off? Depending on the answer, up to $4,000 billion could disappear on the US stock market – i.e. $4 trillion. That’s why the “Economist” headlines about a “$4 trillion accounting puzzle at the heart of artificial intelligence.”
The 4 trillion question arises because such huge amounts of money are being used in the AI boom. Billions are not enough. It’s trillions. In 2024 alone it was almost $1 trillion, estimates the research firm Gartner. By 2026 there will be 2 trillion. A large part of this comes from the five American AI giants: Microsoft, Amazon, Meta, Alphabet and Oracle.
The trillions will be used to buy the latest superchips and create data centers. Everything is huge, visionary, groundbreaking. And yet it has to be accounted for and written off. Like any other investment. Like an oven or a refrigerator in a restaurant.
For superchips, the pace of depreciation depends on how quickly new generations come onto the market. The manufacturer Nvidia wants to release better chips every year. So will what is new today be outdated in a year? What is it worth then? Can you still make money with it? Or will all customers flee to the competition, where the chips are state-of-the-art and the AI works better?
This raises the question for accounting at the AI giants: Is six years appropriate for a superchip? This is how companies currently handle it. Or should the chips be depreciated to zero after three years? After two? After one?
Write-downs as the most hated word in the stock market
The answer depends on what profit the five American AI giants are allowed to report. According to calculations by The Economist, they have the data centers on the books at an estimated value of $156 billion and want to write them off over six years. In 2024 they therefore had to make write-offs of around 26 billion. Your profit was smaller by this 26 billion.
What happens if the giants have to write off their chips after three years? Then twice as much depreciation occurs in these years, which pushes profits down. In 2024, the AI giants would not have written off 26 billion, but 52 billion – and would have made another 26 billion less profit. For these giants, however, that is only 8 percent of the annual profit. But the calculation doesn’t end there either. The real bang only comes on the stock market.
Because every dollar of profit that the AI giants earn is worth many times over on the stock market: 30 times as much. To put it simply: if the AI giants make one dollar more profit, their market value increases by $30. Conversely, this means that one dollar less in profits is penalized 30-fold by the stock market. $26 billion less profit means a $780 billion deduction on the stock market.
If everything has to be written off even faster, i.e. within two years, that would be 52 billion less profit and 1,560 billion less on the stock market. If things were really brutal and the corporations had to write off the entire 156 billion in one year, they would make 130 billion less in profit. On the stock market, $3,900 billion would then disappear into thin air – or almost $4 trillion.
So much is at stake in the dry accounting question of how to depreciate data centers. The “Yahoo Finance” portal therefore writes that “depreciation” is currently the most unpopular word on the stock market.
Famous investors get involved. If the Facebook group Meta actually has to write off its super chips more quickly, then “most of its profits have been significantly overstated,” warns Jim Chanos, who bet against the once-hyped German payment service provider Wirecard – and was spectacularly right.
Michael Burry is even more famous. Before the financial crisis of 2008, he bet on falling real estate prices and was immortalized in the book and then the film “The Big Short”. Burry is now accusing the chip giant Nvidia of “artificially increasing” its profits with excessive depreciation. This approach is “one of the most common types of fraud of our time”.
Superchips like cars that run a 24-hour race every day
Perhaps the question of the value of superchips will soon be answered by the chips themselves. Because it’s not just about the speed at which new generations come onto the market. Equally important, if not more important, is wear and tear, says Paul Kedrosky, a researcher at the Massachusetts Institute of Technology and partner in a venture fund. In an interview with the Bloomberg news agency, Kedrosky warns: Superchips are used very differently than ordinary chips and therefore wear out much more quickly.
They still like him on the stock market: US President Donald Trump. image: Michael Nagle / Bloomberg
Ordinary chips store huge amounts of data. They would be subjected to about as little stress as a car that you would only use to drive to church every Sunday. The wear and tear of such chips is low and, according to Kedrosky, a shelf life of six years is quite possible.
Superchips, on the other hand, are used in a completely different way, says Kedrosky. If you use current Nvidia chips to train an AI model, these chips run around the clock, always under full load, at high temperatures and with huge consumption of electricity and water. With superchips it is like a car that drives the 24 Hours of Le Mans every day, says Kedrosky. Their lifespan is probably only two years, maybe even just 18 months. Such a short lifespan would probably be enough to cause quite an earthquake on the stock market.